When AI seats don't equal AI usage

Who this is for

This playbook is for owner-operated companies that already bought seats — Microsoft, Google, a specialist copilot, a meeting notetaker — and cannot honestly say the work changed. It is also for the operations lead who is about to renew an annual because “we should be using AI” and has no list of who actually does.

If the bill is the loud pain and the jobs were never named, read cut AI spend without a quality cliff next. If the stack is full of overlapping products, not just unused AI, start at overlapping SaaS. If people cannot find a trusted source to work from, seats will not save them — that is operational knowledge.

Symptoms

If several of these are true in the same month, this is a live operational leak — not a tooling preference.

• Licenses were issued company-wide after a kickoff and most people opened the tool once.
• Training was a webinar on a vendor dataset, then everyone went back to the old path.
• Managers report “we have AI” and cannot name a workflow that uses it this week.
• A few enthusiasts have personal prompts; the rest of the team treats the seat as wallpaper.
• Renewal is next month and nobody will say out loud which seats are cold.
• A second AI tool was added because the first one “did not stick,” and both are still on the bill.
• Someone wants an adoption dashboard of logins instead of a list of finished jobs.

What done looks like

Done is a working operating change, not a purchased seat or a dashboard nobody opens.

• Each paid AI seat is tied to a named workflow or is marked for reclaim.
• A 90-day plan exists: jobs, owners, coaching cadence, and a definition of used.
• Training happens on the team’s live work, with the real files and the real review rule.
• Usage is reviewed as completed workflows, not as seat logins.
• Cold seats are reclaimed or paused before the renewal decision.
• A leftover seat that earned renewal has an owner who will keep coaching it.

Seats are inventory. Usage is a job.

Vendors sell enablement. What you needed was a job change. A seat that can draft, summarize, or search is not a workflow until someone uses it on a repeating path: the weekly brief, the first-pass proposal, the meeting note that becomes a task list, the inbox triage that a human still sends. If you cannot name the path, you bought inventory.

“Used” has to be boring and visible. Logged in last week is not used. Generated a novelty poem is not used. Used means the output entered the real path — a draft the reviewer touched, a note the CRM accepted, a brief the meeting actually read. Write that definition down so a login dashboard cannot replace it at renewal.

Enthusiasts are not adoption. One person with a clever prompt library can make the company feel modern in a slide. The test is whether a second person can complete the same job without asking the enthusiast to hover.

Write a 90-day workflow plan

Ninety days is long enough to see a habit and short enough that a bad purchase can still be cut before the year renews. Do not write a transformation program. Write a plan with four columns: workflow, people, coaching owner, and what done looks like by day ninety.

Pick few workflows. Two or three is plenty. Choose work that already happens on a clock: a report, a follow-up, a proposal draft, a support first pass. Do not pick “be more AI-native.” That is not a job.

Name a coaching owner who already sits in the work. Enablement from a vendor success manager cannot see your files or your review rule. An internal owner can sit next to one person, run the real job, and write down what broke. If nobody will own coaching, you do not have a rollout. You have a hope.

Put a midpoint check at thirty days. If nobody completed the workflow twice, stop adding seats. Change the path or reclaim the licenses. Adding more seats to a dead workflow is how the pile grows.

Train in the job, not in demo theater

Demo theater is a polished recording, a sample company, and a prompt that works because the data was cleaned. People nod. Monday they open their own messy folder and the tool looks useless. Train on the live job: their last proposal, their actual inbox, their weekly pack, their real permission constraints.

Training is a reviewed repetition, not a lunch and learn. Sit through one complete cycle. Write the steps the human still owns. Write the steps the tool may draft. Write what must never auto- send. Then have a second person run the same cycle without the first person in the room. If they cannot, the “training” was a show.

Knowledge matters here. If the current policy, rate card, or process lives in folklore, the model will draft from whoever yelled last. Build an operational knowledge base your team can trust before you ask seats to speak for the company. Adoption on top of stale decks produces confident wrong drafts and a team that quietly goes back to the old path.

Reclaim cold seats before renewal

Renewal is where unused inventory becomes next year’s baseline. Before the date, list seats that never met the “used” definition. Pause or remove them. Talk to the few people who did use the tool and ask whether they need the leftover seats or a smaller plan. This is ordinary subscription hygiene. It feels awkward only because AI was sold as inevitable.

Reclaim is not punishment. People who never got a workflow should not keep a license to make the adoption chart look fuller. People who got a workflow and still will not run it can lose the seat until the job is real. You can always add a seat back. You cannot unspend an annual you already signed.

If two AI products were bought because the first “did not stick,” do not renew both out of indecision. That is overlapping SaaS. Pick a keeper for the job or pick neither. A second copilot is not a strategy.

A worked 90 days: meeting notes to next steps

The company already pays for a meeting notetaker on every seat. Most people tried it in week one. A few still open the transcripts. Nobody can say the notes changed a job. Name the workflow: after a client call, a next-step list lands in the place the team already uses, reviewed by the person who was on the call, within a day. That is used. A transcript sitting in the notetaker is not used.

Pick five people who already run those calls. Train on last week’s real recording, not a vendor sample. Write the human steps: confirm the next step, remove anything the model invented, put the rest in the tracker. Meet at day thirty. If three of the five completed the path twice, keep coaching the other two. If nobody did, the workflow is wrong or the coaching owner was a fiction — do not buy a second notetaker.

At day ninety, reclaim the seats that never produced a reviewed next-step list. Renew only for the people who still run the path. That is adoption. It is also how you avoid paying for a company-wide icon that nobody will defend in a renewal meeting.

What people usually get wrong

They measure logins and call it culture. Logins are easy to game and easy to buy. Workflow completion is harder and is the only number that should survive a renewal conversation — and even then, only if you actually counted it. Do not invent an adoption percentage.

They train the whole company at once. A company-wide seat drop feels decisive and produces a company-wide shrug. Start with the people who already own the target job.

They confuse personal enthusiasm with an operating path. A clever user is a gift. They are not a rollout. Write the path so it survives their vacation.

They skip ranking and buy another tool because adoption failed. Sometimes the job was wrong. Read rank AI work that pays before you replace a cold seat with a different logo.

DIY vs hire

DIY the 90-day plan on one workflow. Pick the job that already hurts. Name five people who do it. Define used. Train on last week’s real files. Meet at thirty days. Reclaim anyone still cold before you talk to the vendor about next year. An owner or ops lead can run that without a change-management firm.

Hire help when seats span several departments and nobody will own coaching, when a renewal is days away and the company cannot list usage, or when a second AI purchase is about to happen because the first one is idle. The $997 assessment is a live consultation and a written analysis. You leave with two to four improvements you can run yourself — a workflow list, a used definition, a reclaim pass — and a later roadmap quoted only if you want us on training design or a harder rollout. Implementation is not in the fee.

Controls before automation

Recommendations that touch customers, money, contracts, private data, or systems of record need human review, limited permissions, a log, and a fallback. Do not automate a broken path because the tool is ready.

• Do not auto-send customer, money, or contract drafts produced in training or in production.
• Do not widen permissions “so the copilot can see everything.”
• Do not treat vendor login reports as proof the workflow changed.
• Reclaim or pause cold seats before signing an annual.
• Keep a fallback path that works when the model or the vendor is down.
• If the source documents disagree, stop asking the seat to answer. Fix the source first.

How this sits next to the other playbooks

Adoption is the waste-and-adopt room in the fourteen mid-market AI asks: stop paying for unused capability, and stop calling a purchase a change. Cost reduction is what you do after you know which seats never earned a job. SaaS consolidation is the same map for the rest of the stack. Knowledge is the floor under any draft. Ranking decides whether you even needed the seats.

If you want that plan written with you before renewal week, book the assessment. Bring the license list and one honest workflow. Leave with a 90-day path and a reclaim list.